src/gallium/drivers/panfrost/pan_mfbd.c

   1 /*
   2  * Copyright 2018-2019 Alyssa Rosenzweig
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  */
  24
  25 #include "pan_context.h"
  26 #include "pan_util.h"
  27 #include "pan_format.h"
  28
  29 #include "util/u_format.h"
  30
  31 static struct mali_rt_format
  32 panfrost_mfbd_format(struct pipe_surface *surf)
  33 {
  34         /* Explode details on the format */
  35
  36         const struct util_format_description *desc =
  37                 util_format_description(surf->texture->format);
  38
  39         /* Fill in accordingly */
  40
  41         struct mali_rt_format fmt = {
  42                 .unk1 = 0x4000000,
  43                 .unk2 = 0x1,
  44                 .nr_channels = MALI_POSITIVE(desc->nr_channels),
  45                 .flags = 0x444,
  46                 .swizzle = panfrost_translate_swizzle_4(desc->swizzle),
  47                 .unk4 = 0x8
  48         };
  49
  50         return fmt;
  51 }
  52
  53
  54 static void
  55 panfrost_mfbd_clear(
  56                 struct panfrost_job *job,
  57                 struct bifrost_framebuffer *fb,
  58                 struct bifrost_fb_extra *fbx,
  59                 struct bifrost_render_target *rt)
  60 {
  61         if (job->clear & PIPE_CLEAR_COLOR) {
  62                 rt->clear_color_1 = job->clear_color;
  63                 rt->clear_color_2 = job->clear_color;
  64                 rt->clear_color_3 = job->clear_color;
  65                 rt->clear_color_4 = job->clear_color;
  66         }
  67
  68         if (job->clear & PIPE_CLEAR_DEPTH) {
  69                 fb->clear_depth = job->clear_depth;
  70         }
  71
  72         if (job->clear & PIPE_CLEAR_STENCIL) {
  73                 fb->clear_stencil = job->clear_stencil;
  74         }
  75 }
  76
  77 static void
  78 panfrost_mfbd_set_cbuf(
  79                 struct bifrost_render_target *rt,
  80                 struct pipe_surface *surf,
  81                 bool flip_y)
  82 {
  83         struct panfrost_resource *rsrc = pan_resource(surf->texture);
  84         int stride = rsrc->bo->slices[0].stride;
  85
  86         rt->format = panfrost_mfbd_format(surf);
  87
  88         /* Now, we set the layout specific pieces */
  89
  90         if (rsrc->bo->layout == PAN_LINEAR) {
  91                 mali_ptr framebuffer = rsrc->bo->gpu;
  92
  93                 if (flip_y) {
  94                         framebuffer += stride * (surf->texture->height0 - 1);
  95                         stride = -stride;
  96                 }
  97
  98                 /* MFBD specifies stride in tiles */
  99                 rt->framebuffer = framebuffer;
 100                 rt->framebuffer_stride = stride / 16;
 101         } else if (rsrc->bo->layout == PAN_AFBC) {
 102                 rt->afbc.metadata = rsrc->bo->afbc_slab.gpu;
 103                 rt->afbc.stride = 0;
 104                 rt->afbc.unk = 0x30009;
 105
 106                 rt->format.flags |= MALI_MFBD_FORMAT_AFBC;
 107
 108                 mali_ptr afbc_main = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 109                 rt->framebuffer = afbc_main;
 110
 111                 /* TODO: Investigate shift */
 112                 rt->framebuffer_stride = stride << 1;
 113         } else {
 114                 fprintf(stderr, "Invalid render layout (cbuf)");
 115                 assert(0);
 116         }
 117 }
 118
 119 static void
 120 panfrost_mfbd_set_zsbuf(
 121                 struct bifrost_framebuffer *fb,
 122                 struct bifrost_fb_extra *fbx,
 123                 struct pipe_surface *surf)
 124 {
 125         struct panfrost_resource *rsrc = pan_resource(surf->texture);
 126
 127         if (rsrc->bo->layout == PAN_AFBC) {
 128                 fb->unk3 |= MALI_MFBD_EXTRA;
 129
 130                 fbx->flags =
 131                         MALI_EXTRA_PRESENT |
 132                         MALI_EXTRA_AFBC |
 133                         MALI_EXTRA_AFBC_ZS |
 134                         MALI_EXTRA_ZS |
 135                         0x1; /* unknown */
 136
 137                 fbx->ds_afbc.depth_stencil_afbc_metadata = rsrc->bo->afbc_slab.gpu;
 138                 fbx->ds_afbc.depth_stencil_afbc_stride = 0;
 139
 140                 fbx->ds_afbc.depth_stencil = rsrc->bo->afbc_slab.gpu + rsrc->bo->afbc_metadata_size;
 141
 142                 fbx->ds_afbc.zero1 = 0x10009;
 143                 fbx->ds_afbc.padding = 0x1000;
 144         } else if (rsrc->bo->layout == PAN_LINEAR) {
 145                 fb->unk3 |= MALI_MFBD_EXTRA;
 146                 fbx->flags |= MALI_EXTRA_PRESENT | MALI_EXTRA_ZS | 0x1;
 147
 148                 fbx->ds_linear.depth = rsrc->bo->gpu;
 149                 fbx->ds_linear.depth_stride = rsrc->bo->slices[0].stride;
 150         } else {
 151                 assert(0);
 152         }
 153 }
 154
 155 /* Helper for sequential uploads used for MFBD */
 156
 157 #define UPLOAD(dest, offset, src, max) { \
 158         size_t sz = sizeof(*src); \
 159         memcpy(dest.cpu + offset, src, sz); \
 160         assert((offset + sz) <= max); \
 161         offset += sz; \
 162 }
 163
 164 static mali_ptr
 165 panfrost_mfbd_upload(
 166                 struct panfrost_context *ctx,
 167                 struct bifrost_framebuffer *fb,
 168                 struct bifrost_fb_extra *fbx,
 169                 struct bifrost_render_target *rts,
 170                 unsigned cbufs)
 171 {
 172         off_t offset = 0;
 173
 174         /* There may be extra data stuck in the middle */
 175         bool has_extra = fb->unk3 & MALI_MFBD_EXTRA;
 176
 177         /* Compute total size for transfer */
 178
 179         size_t total_sz =
 180                 sizeof(struct bifrost_framebuffer) +
 181                 (has_extra ? sizeof(struct bifrost_fb_extra) : 0) +
 182                 sizeof(struct bifrost_render_target) * cbufs;
 183
 184         struct panfrost_transfer m_f_trans =
 185                 panfrost_allocate_transient(ctx, total_sz);
 186
 187         /* Do the transfer */
 188
 189         UPLOAD(m_f_trans, offset, fb, total_sz);
 190
 191         if (has_extra)
 192                 UPLOAD(m_f_trans, offset, fbx, total_sz);
 193
 194         for (unsigned c = 0; c < cbufs; ++c) {
 195                 UPLOAD(m_f_trans, offset, &rts[c], total_sz);
 196         }
 197
 198         /* Return pointer suitable for the fragment section */
 199         return m_f_trans.gpu | MALI_MFBD | (has_extra ? 2 : 0);
 200 }
 201
 202 #undef UPLOAD
 203
 204 /* Creates an MFBD for the FRAGMENT section of the bound framebuffer */
 205
 206 mali_ptr
 207 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool flip_y)
 208 {
 209         struct panfrost_job *job = panfrost_get_job_for_fbo(ctx);
 210
 211         struct bifrost_framebuffer fb = panfrost_emit_mfbd(ctx);
 212         struct bifrost_fb_extra fbx = {};
 213         struct bifrost_render_target rts[4] = {};
 214
 215         /* XXX: MRT case */
 216         fb.rt_count_2 = 1;
 217         fb.unk3 = 0x100;
 218
 219         /* TODO: MRT clear */
 220         panfrost_mfbd_clear(job, &fb, &fbx, &rts[0]);
 221
 222         for (int cb = 0; cb < ctx->pipe_framebuffer.nr_cbufs; ++cb) {
 223                 struct pipe_surface *surf = ctx->pipe_framebuffer.cbufs[cb];
 224                 panfrost_mfbd_set_cbuf(&rts[cb], surf, flip_y);
 225         }
 226
 227         if (ctx->pipe_framebuffer.zsbuf) {
 228                 panfrost_mfbd_set_zsbuf(&fb, &fbx, ctx->pipe_framebuffer.zsbuf);
 229         }
 230
 231         /* For the special case of a depth-only FBO, we need to attach a dummy render target */
 232
 233         if (ctx->pipe_framebuffer.nr_cbufs == 0) {
 234                 struct mali_rt_format null_rt = {
 235                         .unk1 = 0x4000000,
 236                         .unk4 = 0x8
 237                 };
 238
 239                 rts[0].format = null_rt;
 240                 rts[0].framebuffer = 0;
 241                 rts[0].framebuffer_stride = 0;
 242         }
 243
 244         /* When scanning out, the depth buffer is immediately invalidated, so
 245          * we don't need to waste bandwidth writing it out. This can improve
 246          * performance substantially (Z32_UNORM 1080p @ 60fps is 475 MB/s of
 247          * memory bandwidth!).
 248          *
 249          * The exception is ReadPixels, but this is not supported on GLES so we
 250          * can safely ignore it. */
 251
 252         if (panfrost_is_scanout(ctx)) {
 253                 job->requirements &= ~PAN_REQ_DEPTH_WRITE;
 254         }
 255
 256         /* Actualize the requirements */
 257
 258         if (job->requirements & PAN_REQ_MSAA) {
 259                 rts[0].format.flags |= MALI_MFBD_FORMAT_MSAA;
 260
 261                 /* XXX */
 262                 fb.unk1 |= (1 << 4) | (1 << 1);
 263                 fb.rt_count_2 = 4;
 264         }
 265
 266         if (job->requirements & PAN_REQ_DEPTH_WRITE)
 267                 fb.unk3 |= MALI_MFBD_DEPTH_WRITE;
 268
 269         if (ctx->pipe_framebuffer.nr_cbufs == 1) {
 270                 struct panfrost_resource *rsrc = (struct panfrost_resource *) ctx->pipe_framebuffer.cbufs[0]->texture;
 271
 272                 if (rsrc->bo->has_checksum) {
 273                         fb.unk3 |= MALI_MFBD_EXTRA;
 274                         fbx.flags |= MALI_EXTRA_PRESENT;
 275                         fbx.checksum_stride = rsrc->bo->checksum_stride;
 276                         fbx.checksum = rsrc->bo->gpu + rsrc->bo->slices[0].stride * rsrc->base.height0;
 277                 }
 278         }
 279
 280         /* We always upload at least one (dummy) cbuf */
 281         unsigned cbufs = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
 282
 283         return panfrost_mfbd_upload(ctx, &fb, &fbx, rts, cbufs);
 284 }